Cloud Computing and the Future of Internet Services. Wei-Ying Ma Principal Researcher, Research Area Manager Microsoft Research Asia

Transcription

1 Cloud Computing and the Future of Internet Services Wei-Ying Ma Principal Researcher, Research Area Manager Microsoft Research Asia

2 Computing as Utility Grid Computing Web Services in the Cloud What is Cloud Computing? Software as a Service (SaaS) Platform as a Service (PaaS) 雲 計 算 雲 端 計 算 網 雲 計 算 Sky Computing

3 Comparison Grid Computing Refer to resource-pooled environments for running compute jobs (like image processing) rather than long running processes (such as a Web site or server) Utility Computing Refer to resource-pooled environments for hosting long running processes, and tends to be focused on meeting service levels with the optimal amount of resources necessary to do so Cloud Computing Refer to a variety of services available over the Internet that deliver compute functionality on the service provider's infrastructure Its environment (infrastructure) may actually be hosted on either a grid or utility computing environment, but that doesn't matter to a service user The data in the cloud, as Intel inside (or intelligence inside), is often an important part of the services

4 Cloud Computing My Simple Definition Cloud Computing = Software as a Service + Platform as a Service + Data as a Service + Infrastructure as a Service

5 Two Important Aspects for Cloud Computing Operate Internet-based services in the cloud Host and run high-level web apps such Hotmail, Search, Virtual Earth, Office Online, Xbox Live software as a service Offer an Internet-based platform to developers who want to create services but don't have their own cloud to run them on Rent storage, computation, and maintenance (datacenters) from someone else platform as a service and infrastructure as a service

6 Cloud Computing Software as a Service (SaaS) From user s point of view Apps are located in the cloud (i.e. somewhere in the Web) Software experiences are delivered through the Internet

7 Cloud Computing Platform as a Service (PaaS) From developer s point of view Similar to Windows Platform for ISV developers in the PC ecosystem Computing as utility Cloud computing providers builds and maintains the infrastructure and provides the cloud platform for developers to build applications and end-user services Companies that build end-user services can buys time/resources on this infrastructure

8 A Hypothetical Scenario The Next Facebook What platform will be chosen by three Harvard students for their internet service idea? They will need to Design web pages Develop applications (database, business logic) Find contents and users Application Buy machines Find a place to host machines Network Platform issues

9 A Service to Solve Platform Issues Should be cheap and easy to get up and running Should provide all essential building blocks e.g., computation, structured storage, network, load balancing, service management Should scale seamlessly e.g., storage capacity, blob size, bandwidth, computation Should be highly available 99.99% availability for all components Should provide rich development environment Visual Studio for Services that provides end-to-end developer support

10 The data in the cloud Cloud Computing The large amount of data available on the web form the engine for new Web applications and Internet services The meshes created by various services and applications further interconnect the data These data form intelligence inside for new Web applications and services Data as a Service Data is a strategic asset for a company

11 Cloud Computing Infrastructure as a Service Build datacenters Include power, scale, hardware, networking, storage, distributed systems, etc Distributed data management and geo-distribution for locality, scalability, load balancing, and disaster tolerance Support web-scale offline and real-time computation Companies that build cloud services can buy time/resources on the infrastructure (like utility) Data-intensive computing becomes a commodity that enables more data-intensive business

12 Microsoft s Mega-Datacenters

13 Microsoft s Cloud Platform Applications & Solutions Live Platform Services Cloud Infrastructure Services This layer does on-demand utility computing back end hooks up with rich Internet applications on the front end, e.g. Office Live CRM, Windows Live services, Xbox Live, MSN, Live Search This layer provides services designed specifically to serve the needs of applications, e.g. identity services, contact lists, social graph, communication services, and advertising platform This layer provides Virtualization support for all online services to run on Application frameworks designed to support a variety of application models Automatic deployment, load balancing, performance optimization, and horizontally scalable storage including searchable storage functionality Global Foundation Services This layer contains physical data centers along with the networks that connect them

14 Two Major Types of Computing Computation-intensive computing Focus on raw computing resource in the cloud Computing as utility and paid for usage Data-intensive computing Data as an important part of the services Need to manage and process lots of data E.g. Search, Advertising, Mapping, etc

15 Search and Data-intensive Computing Search leads the development of infrastructure for data-intensive computing Deep analysis and integration of web data Intelligence inside for future web applications & services Search Web Data Intelligence Advertising Audience Intelligence Future Web Applications & Services Infrastructure for Internet-scale Data Management Collect Store Aggregate Analyze Present Act

16 Traditional Search Engine Pipeline Fetch (Crawl) Process Build (Index) Serve Guarantee the smooth running of the pipeline Highly optimized for performance and reliability

17 Problems and Limitations Limited to light-weight, freshness-critical data processing -> Shallow analysis Lack of data management capability -> Transient and scattered data Data is second class citizen Difficult to accumulate knowledge (meta-data) Tightly coupled system -> Less flexibility &extensibility A small change in one place could create a ripple effect in the entire system

18 Why Search Infrastructure Provide Web data management for search Highly optimized for flexibility & extensibility Allow expensive, asynchronous data processing Support deep analysis and understanding Web documents Queries Users

19 Experimental Infrastructure Search needs continuous experiments An infrastructure for scale experiments is critical Three systems Offline experiments (-> offline computation) Online experiments (-> real-time computation) Production (-> computing as utility)

20 Infrastructure for Data-intensive Computing Application Script Language Function Easy-to-use Data Processing Tools Control API Distributed Programming and Execution Distributed Programming API Massive Storage Storage System API Data Acquisition

21 Web-Scale Data Mining for Search Fetch (Crawl) Process Build (Index) Serve Data Exchange Data Exchange Data Exchange Web-Scale Data Mining Infrastructure Classification / Clustering Information Extraction Document Features Link-based Features Entity/Object Extraction Entity Relationship Mining Statistics Vertical / Multimedia / Mobile Experimentations Expensive Offline Computations Massive Store

22 Renlifang ( 人 立 方 ) Search: Renlifang: Mining relationships among people, organizations and locations from crawled Web pages Covers head and tail queries On both regular and well-known people, locations, and organizations Key technologies Entity Extraction and Summarization Expert Finding (based on Surrounding Text) Entity Relationship Mining and Exploration (based on Web cooccurrence)

23 Search is a Driving Force for Cloud Computing Build Infrastructure for data-intensive computing Datacenters and layered infrastructure Web-scale data mining & machine learning Understand, extract, and expose semantics for future web services and applications Deep integration of Web data Data as intelligence inside Killer App for cloud platform to drive Internet economies on a global scale

24 Cloud Computing and Internet Services Microsoft provides the cloud platform for developers to build Internet services End-User Services Search (Killer App) Information Content Social networks (SNS) Productivity Services (Office Live) Communication (IM) Subscription Services Online Commerce (C2C, B2C, B2B, etc) Monetization Platform Advertising (Buy/Sell/Serve) Payment/Marketplace Audience Intelligence Data/Business Intelligence Cloud Platforms (Platform as a Service & Computing as Utility) Cloud Application Services Cloud Infrastructure Services Cloud Foundation Services

25 Datacenter Evolution Generation 1: Traditional data centers that focus on uptime, reliability, and redundancy Generation 2: Built with more energy efficiency and sustainability in mind Generation 3: Built for massive scalability and cost efficiency Problems and Challenges Need to build data centers to be just right in size in order to squeeze out the most energy efficiency The worst thing we can do in delivering facilities for the business is not have enough capacity online, thus limiting the growth of our products and services The second worst thing we can do in delivering facilities for the business is to have too much capacity online

26 Microsoft s Gen 4 Modular Datacenter Generation 4: Modular datacenter and smart growth Commoditize the build-out of data centers, assembling data center components on-site Use shipping containers filled with servers as their basic building block Modular units of prefabricated mechanical, electrical, and security components A container facility helps ensure that we don't overbuild server capacity, while allowing the company to reduce the time to build a data center Decrease the need for water chillers by using ambient air to cool data centers

27 Microsoft s Gen 4 Modular Datacenter Video:

28 Summary The real underlying value of cloud computing is that it transparently makes software and data available everywhere Computing as utility and data-intensive business Cloud computing will have profound impacts on Internet economies Intelligence inside for Internet economies (data driven) Speed of light (algorithms driven) Scale to global (infrastructure driven)

29 Thanks You!